Integrating multi-domain deep features of electrocardiogram and phonocardiogram for coronary artery disease detection

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Electrocardiogram (ECG) and phonocardiogram (PCG) are noninvasive tools for detecting heart abnormalities like coronary artery disease (CAD).
• Diagnosing CAD using only ECG or PCG presents challenges due to low sensitivity.
• Previous attempts to combine ECG and PCG for heart abnormality diagnosis relied on conventional manual features.

Purpose of the Study:

• To develop a deep learning framework for enhanced coronary artery disease (CAD) detection by integrating multi-domain features from ECG and PCG signals.
• To leverage the feature extraction capabilities of Convolutional Neural Networks (CNNs) for improved diagnostic accuracy in CAD.
• To assess the effectiveness of a multi-input CNN framework in classifying subjects with and without CAD.

Main Methods:

• A multi-input Convolutional Neural Network (CNN) framework was developed, integrating deep features from time, frequency, and time-frequency domains of ECG and PCG signals.
• The framework utilized both 1-D and 2-D CNN models, processing raw signals, spectrum images, and time-frequency images.
• Simultaneously recorded ECG and PCG data from 195 subjects were used for training and validation.

Main Results:

• The proposed multi-input CNN framework achieved high diagnostic performance for CAD detection.
• Classification accuracy reached 96.51%, with a sensitivity of 99.37% and a specificity of 90.08%.
• The method demonstrated competitive results compared to existing CAD detection studies.

Conclusions:

• The developed deep learning framework effectively integrates multi-domain features from ECG and PCG for accurate CAD detection.
• This approach shows significant promise in assisting real-world clinical diagnosis of CAD, particularly under general medical conditions.
• The study highlights the potential of deep learning in enhancing the diagnostic capabilities of noninvasive cardiovascular monitoring tools.